Secondary Index Search in InnoDB
0 likes254 views
The document discusses secondary index searches in MySQL. It describes the process as starting with a search of the secondary index tree to find the primary key. The primary key is then added to an unsorted list. Once all secondary index searches are complete, the primary key list is sorted. The primary index tree is then searched sequentially using the sorted primary key list to retrieve the clustered data records. Finally, the clustered data records are accessed sequentially.
![Reference
8/23/16 25
[1] “MySQL 5.6 Reference Manual”, MySQL, https://dev.mysql.com/doc/refman/5.6/en/
[2] Jeremy Cole, “InnoDB”, https://blog.jcole.us/innodb/](https://image.slidesharecdn.com/secondary-index-search-210324065446/85/Secondary-Index-Search-in-InnoDB-25-320.jpg)
Secondary Index Search in InnoDB
- 1. Secondary Index Search in MySQL Mijin An [email protected] 8/23/16 1
- 2. Contents 8/23/16 2 • Secondary/Primary Key Record Format • Secondary Index Search • Proposal for Improvement • Improved Secondary Index Search
- 3. Secondary Key Record Format 8/23/16 3 = Primary Key Value
- 4. Primary Key Record Format 8/23/16 4
- 5. Secondary Index Search 8/23/16 5 ① Start a mini-transaction ② Searches an index tree, then positions a tree cursor(pcur) on a record corresponding to the index (btr_pcur_open_with_no_init()) ③ Get the record that pcur indicates(rec) ④ Go to rec_loop: row_search_for_mysql()
- 6. rec_loop: 8/23/16 6 a. Get the record that pcur indicates(rec) b. If record is infimum or supremum, go to next_rec: Look for matching records in a loop
- 7. next_rec: 8/23/16 7 a. Stores current position of pcur b. Commit mini-transaction c. If record is infimum è move pcur to the next record Else if record is supremum è move pcur to the first record of next page Else è move pcur according to the search direction d. If we may need to process the record the cursor is now on, go to rec_loop: Move cursor to the next record, then search again
- 8. rec_loop: 8/23/16 8 a. Get the record that pcur indicates(rec) b. If record is infimum or supremum, go to next_rec: c. Caculates the offsets to each field in the record (offsets) d. If select_lock_type == LOCK_S or LOCK_X, - try to place a lock on the index record e. Go to locks_ok: If index != clust_index && need_to_access_clustered, - go to requires_clust_rec: Look for matching records in a loop
- 9. requires_clust_rec: 8/23/16 9 a. Retrieves the clustered record(clust_rec) corresponding to a record in a secondary index (row_sel_get_clust_rec_for_mysql()) Get clustered record using secondary index
- 10. row_sel_get_clust_rec_for_mysql() 8/23/16 10 a. Get the clust_index from sec_index b. Searches an index tree, then positions a tree cursor (clust_pcur) on a record corresponding to the index (btr_pcur_open_with_no_init()) c. Get the record that clust_pcur indicates (clust_rec) d. Caculate the offsets to each field in the record (offsets) e. If select_lock_type == LOCK_S or LOCK_X, - try to place a lock on the index record Get clustered record using secondary index
- 11. row_sel_get_clust_rec_for_mysql() 8/23/16 11 f. Go to func_exit: g. In func_exit:, store the current clust_pcur position if select_lock_type is LOCK_S or LOCK_X Get clustered record using secondary index
- 12. requires_clust_rec: 8/23/16 12 a. Retrieves the clustered record(clust_rec) corresponding to a record in a secondary index (row_sel_get_clust_rec_for_mysql()) b. If clust_rec is delete marked, skip it è next_rec: c. Store clust_rec into the result_rec d. Go to normal_return: Get clustered record using secondary index
- 13. Secondary Index Search 8/23/16 13 ⑤ Commit mini-transaction ⑥ Go to func_exit: ⑦ In func_exit:, frees the space occupied by a memory heap then exit row_search_for_mysql() è normal_return: è func_exit:
- 14. Data records Secondary Index Search 8/23/16 14 Primary key Secondary index tree Primary index tree Primary index tree Data records
- 15. Data records Secondary Index Search 8/23/16 15 Primary key Secondary index tree Primary index tree Primary index tree Data records
- 16. Data records Secondary Index Search 8/23/16 16 Primary key Secondary index tree Primary index tree Primary index tree Data records
- 17. Data records Proposal: Prefetching in MySQL 8/23/16 17 Primary key Secondary index tree Primary index tree P1 Primary key list(unsorted)
- 18. Data records 8/23/16 18 Primary key Secondary index tree Primary index tree P1 P2 Primary key list(unsorted) Proposal: Prefetching in MySQL
- 19. Data records 8/23/16 19 Primary key Secondary index tree Primary index tree P1 P2 P3 Primary key list(unsorted) Proposal: Prefetching in MySQL
- 20. P2 P3 Data records 8/23/16 20 Primary key Secondary index tree Primary index tree P1 P2 P3 Primary key list(unsorted) Primary key list(sorted) Proposal: Prefetching in MySQL
- 21. P2 P3 Data records 8/23/16 21 Secondary index tree Primary index tree P1 Primary key list(sorted) Primary index tree Primary index tree Data records Data records Proposal: Prefetching in MySQL
- 22. P2 P3 Data records Proposal: Prefetching in MySQL 8/23/16 22 Secondary index tree Primary index tree P1 Primary key list(sorted) Primary index tree Primary index tree Data records Data records
- 23. P2 P3 Data records Proposal: Prefetching in MySQL 8/23/16 23 Secondary index tree Primary index tree P1 Primary key list(sorted) Primary index tree Primary index tree Data records Data records
- 24. Improved Secondary Index Search 8/23/16 24 ① Search a data record using secondary index ② Get primary key and insert it into primary key list, then repeat ①~② process until there is no record to search using secondary index ③ Sort the primary key list ④ Search primary index tree sequentially using sorted primary key list ⑤ Access data record sequentially
- 25. Reference 8/23/16 25 [1] “MySQL 5.6 Reference Manual”, MySQL, https://dev.mysql.com/doc/refman/5.6/en/ [2] Jeremy Cole, “InnoDB”, https://blog.jcole.us/innodb/